Retrieving Seasonal Disaster Records from Early-19th-Century Diaries

Abstract

Disaster records retrieved from historical diaries are valuable for examining past seasonal variations in disaster occurrence. We extracted 154 fire and 103 flood records between 1807 and 1838 from the Zayu-Nichiroku (the Kakuson Diaries), written by KANEKO Kakuson in Kanazawa, Japan. We analyzed the seasonal probability of these events using the Poisson distribution. The probability of fire peaked between March and June, while that of floods was highest in June, July, and September. These trends align well with the current climate in Kanazawa, where low humidity and strong winds elevate fire risk, while prolonged rainfall and localized heavy precipitation during the rainy and typhoon seasons increase flood risk. However, extracting disaster records from historical diaries involves uncertainties stemming from omitted entries, the loss of archival material, ambiguous descriptions, unique local recording bias, and short-term missing records. To reduce these uncertainties, we should employ an interdisciplinary approach utilizing multiple historical sources and probabilistic analyses of disaster occurrence.

1. Introduction

Disasters seriously affect ecosystem services and biodiversity [1,2,3,4]. To accurately evaluate these effects, long-term disaster records including specific dates and locations are essential. Modern records, which rely on in situ and satellite observations, statistical data, and online information, are valuable but limited in temporal scope. To understand periods before modern monitoring began (e.g., prior to 150 years ago in Japan), it is necessary to access historical records extracted from historical diaries and documents. This so-called “historical dark data” [5,6] enables the collection of dates and locations for disasters such as fires, floods, earthquakes, tsunamis, volcanic eruptions, droughts, and famines at centennial and millennial scales [7,8,9,10,11,12,13,14,15,16,17].

The temporal impact of earthquakes, tsunamis, and volcanic eruptions on ecosystem services and biodiversity typically spans decades to centuries, whereas the impact of fires and floods generally ranges from days to years. The spatial extent of the effect of fires and floods (10 m to 10 km scales) is much narrower than that of earthquakes, tsunamis, and volcanic eruptions (10 km to 100 km). Excluding travel accounts or hearsay, the spatiotemporal extent of old diaries and documents written by a single author is usually limited to the writer’s immediate living and activity area (ranging from a few to several dozen kilometers) over a relatively short period (a few years to several decades). Consequently, single-author diaries and documents tend to be highly consistent in their recording of specific events. These characteristics suggest that fires and floods are ideal targets for historical data retrieval. Because specific weather conditions trigger these events, a distinct seasonal variation in their occurrence is expected; however, previous studies have not sufficiently examined these seasonal patterns.

To resolve this issue, we clarify (1) the utility of historical dark data for the analysis of seasonal trends in disaster occurrence, and (2) explore the limitations and uncertainties inherent in retrieving disaster data from historical documents. To this end, we extract fire and flood records spanning approximately 30 years in the early 19th century from a diary written by a single author to clarify their seasonal frequency, causal factors, and similarities to or differences from current fire and flood patterns.

2. Materials and Methods

2.1. The Diary Source

This study utilized the Zayu-Nichiroku (hereafter referred to as the “Kakuson Diaries”) written by the Japanese Confucian scholar KANEKO Kakuson (1758–1840) in Kanazawa, Japan (36°33′40″ N 136°39′24″ E). Despite the difficulty of retrieving continuous, centennial-scale records from a single historical source, using a single-author diary minimizes the systematic noise that would be caused by the varying recording styles of multiple authors (e.g., the Sekiguchi Diaries [Yokohama (35°27′02″ N, 139°38′03″ E), 1762–1901] [18] and the Diaries of the Toyama Family [Hachinohe (40°30′44″ N, 141°29′18″ E), 1792–1919] [19]).

During the period covered in the diary, Kanazawa was located in the castle town of the Kaga Domain, known as Kaga Hyakumangoku (referring to its status as a “Million-Koku” region, where a koku is a traditional unit of rice production), and a center of flourishing samurai culture [20]. The Kakuson Diaries contain diverse observations, including daily weather, social events, disasters, meals, and plant phenology. Previous research has utilized these records to reconstruct historical climate patterns [21], analyze food culture [22,23,24], and study plant phenology [25,26].

2.2. Temporal Coverage and Data Extraction

For this analysis, we used the printed editions of the Kakuson Diaries [27,28,29,30,31,32]. The diary entries span the period from 18 August 1807 (15 July, Bunka 4, in the Japanese calendar) to 9 October 1838 (21 August, Tenpo 9). However, several gaps exist within the record, most notably throughout 1811, 1815, and the mid-1830s (see details in

Table 1. Missing periods in the Kakuson Diaries.

Start Date (Gregorian)	End Date (Gregorian)	Japanese Calendar Period
25 January 1811	12 February 1812	1 January Bunka 8–30 December Bunka 8
9 February 1815	28 January 1816	1 January Bunka 12–30 December Bunka 12
31 January 1824	5 February 1824	1 January Bunsei 7–6 January Bunsei 7
13 September 1833	17 January 1834	30 July Tenpo 4–8 December Tenpo 4
9 February 1834	16 February 1836	1 January Tenpo 5–30 December Tenpo 6
7 March 1837	4 April 1837	1 February Tenpo 8–29 February Tenpo 8
13 June 1837	20 September 1837	11 May Tenpo 8–21 August Tenpo 8

; [22,24]). We assumed these missing records did not affect the analysis because there was no extreme seasonal bias.

2.3. Statistical Analysis

Because records from the Kakuson Diaries may not capture every instance of fire or flood in Kanazawa, we analyzed the monthly probability of these events. The probability of fire and flood occurrence was assumed to follow a Poisson distribution, which expresses the probability of a given number of events occurring over a certain fixed interval of time. Suppose an event happens on average λ times in a given period, the probability of an event occurring k times, P(X = k), is defined by Equation (1):

$$P\left(X=k\right)={\displaystyle \frac{e^{-\lambda }{\lambda }^k}{k!}}\left(k=0,1,2,\cdots \right)$$

(1)

where e is Euler’s number.

The expected value (mean) and variance of the Poisson distribution are $E\left(X\right)=\lambda$ and $V\left(X\right)=\lambda$, respectively.

2.4. Data Processing and Quality Control

First, we extracted records of fire and flooding events for the period between 1807 and 1838 in and around Kanazawa. We then applied a quality control (QC) flag to each record in the dataset (see Supplementary_fire_Kakuson_diaries.csv and Supplementary_flood_Kakuson_diaries.csv). A QC flag of 1 was assigned when the location of the event could be verified via historical online databases or geographical records; a QC flag of 0 was applied when no evidence for the specific place name could be confirmed. Despite the lack of direct evidence, contextual information within the diaries supported the conclusion that flag-0 records nonetheless referred to locations in and around Kanazawa. Because the Kakuson Diaries follow the Japanese luni-solar calendar, all dates were converted to the Gregorian calendar by established online conversion tools [33,34]. Finally, we calculated the mean monthly frequency for both fires and floods (λ) and used Equation (1) to determine the probability mass functions for 0 to 5 occurrences per month.

Statistical analyses and calculations were performed using R ver. 4.5.2 [35] within the RStudio environment (ver. 2024.12.1, Build 563) [36]. Data organization and preliminary processing were conducted using LibreOffice ver. 24.2.6.2 [37] and custom shell scripts.

3. Results

3.1. Disaster Frequency and Data Quality

A total of 154 fire and 103 flood records were extracted from the Kakuson Diaries (Figure 1a,c; see Supplementary Materials). Fire records peaked in March, with fewer occurrences documented in February, July, and from September to December (Figure 1b). Fifty fire records (32.5%) were assigned a QC flag of 0 owing to unclear descriptions or defunct place names that could not be verified.

Flood records were most frequent in July, with few records in January, February, May, and from October to December (Figure 1d). Only eight flood records (7.8%) were assigned a QC flag of 0, owing to unclear descriptions. Notably, 92 of the recorded flood events occurred along the Sai River (Saigawa).

3.2. Seasonal Probability Analysis

The monthly probability of disaster occurrence was calculated using the Poisson distribution parameters shown in

Table 2. Monthly Poisson Distribution Statistics for Fire and Flood Events in Kanazawa (1807–1838).

Month	Fire		Flood
	Mean	Variance	Mean	Variance
January	0.52	0.52	0.07	0.07
February	0.36	0.36	0.14	0.14
March	0.96	0.96	0.31	0.31
April	0.65	0.65	0.38	0.38
May	0.62	0.62	0.04	0.04
June	0.58	0.58	0.50	0.50
July	0.36	0.36	1.28	1.28
August	0.54	0.54	0.35	0.35
September	0.22	0.22	0.56	0.56
October	0.38	0.38	0.23	0.23
November	0.36	0.36	0.08	0.08
December	0.36	0.36	0.04	0.04

. For fire events, excluding the probability of zero occurrences, the probability of 1 to 3 events occurring per year was highest in March. The probability of zero fires occurring in March was much lower than in other months (approx. 0.4; Figure 2a).

For floods, the probability of 1 to 3 events occurring per month was higher in June, July, and September compared to the rest of the year. The probability of zero floods in July was the lowest of any month (approx. 0.3; Figure 2b).

4. Discussion

4.1. Seasonal Probability of Fire and Flood Occurrence

The probability of fire and flood events in Kanazawa during the early 19th century shows distinct seasonal patterns (Figure 1 and Figure 2). These historical patterns can be explained by comparing them to the contemporary climate of Kanazawa (Figure 3; [38]). This comparison is based on the premise that while absolute temperatures and precipitation totals may have fluctuated since the early 19th century, the fundamental regional drivers (such as the interaction between local topography and seasonal atmospheric circulation) remain consistent. Consequently, the meteorological mechanisms responsible for disasters today provide a reliable framework for interpreting the patterns recorded in the Kakuson Diaries.

Regarding fires, the high probability observed between March and June corresponds with periods of low humidity and strong winds. Monthly precipitation from March to June is lower than in other months, and relative humidity is lowest between March and May. Furthermore, wind speeds from November through April (the snow and snow melting periods) are higher than in other months. In particular, the peak in fire risk during March and April is exacerbated by the Foehn phenomenon, which occurs in Kanazawa when a migratory anticyclone moves southeast while a low-pressure system or trough moves toward the Sea of Japan, leading to warm air advection on the eastern side of the low [39]. These conditions result in sudden increases in temperature, dryness, and strong winds, which collectively increase the risk of fire.

Conversely, the elevated flood risk in June, July, and September corresponds to periods of prolonged rainfall and localized heavy precipitation. These are associated with the rainy season (June and July) and typhoon season (September) [40]. During these months, heavy rainfall leads to rising river levels, increasing the risk of flooding.

4.2. Comparison of the Early 19th Century and the Present

The climate during the early 19th century coincided with the latter part of the Little Ice Age, characterized by temperatures lower than those of the present day [41,42,43]. To evaluate how these climatic shifts may have influenced the seasonal probabilities of fire and flood occurrence, we calculated the current probability of fire and flood events by using modern statistical data [44,45] (Figure 4; see Appendix A for details).

Figure 4. Seasonal disaster probabilities in present-day Kanazawa. (a,c) Monthly total of observed fire records (2001–2023) and frequency of days with daily precipitation exceeding 50 mm (1995–2024). (b,d) Probability mass functions for fire and daily precipitation exceeding 50 mm by month. Corresponding statistical parameters are summarized in Table 3.

Figure 4. Seasonal disaster probabilities in present-day Kanazawa. (a,c) Monthly total of observed fire records (2001–2023) and frequency of days with daily precipitation exceeding 50 mm (1995–2024). (b,d) Probability mass functions for fire and daily precipitation exceeding 50 mm by month. Corresponding statistical parameters are summarized in Table 3.

Table 3. Monthly Poisson Distribution Statistics for Fire and Heavy Precipitation Events in present-day Kanazawa.

Month	Fire		Daily Precipitation ≥ 50 mm
	Mean	Variance	Mean	Variance
January	7.35	7.35	0.03	0.03
February	6.26	6.26	0.03	0.03
March	9.74	9.74	0.07	0.07
April	11.13	11.13	0.1	0.1
May	9.04	9.04	0.43	0.43
June	7.09	7.09	0.7	0.7
July	6.87	6.87	1.23	1.23
August	6.96	6.96	1.03	1.03
September	7	7	1.03	1.03
October	6.43	6.43	0.43	0.43
November	7.87	7.87	0.7	0.7
December	8.04	8.04	0.6	0.6

Table 3. Monthly Poisson Distribution Statistics for Fire and Heavy Precipitation Events in present-day Kanazawa.

Month	Fire		Daily Precipitation ≥ 50 mm
	Mean	Variance	Mean	Variance
January	7.35	7.35	0.03	0.03
February	6.26	6.26	0.03	0.03
March	9.74	9.74	0.07	0.07
April	11.13	11.13	0.1	0.1
May	9.04	9.04	0.43	0.43
June	7.09	7.09	0.7	0.7
July	6.87	6.87	1.23	1.23
August	6.96	6.96	1.03	1.03
September	7	7	1.03	1.03
October	6.43	6.43	0.43	0.43
November	7.87	7.87	0.7	0.7
December	8.04	8.04	0.6	0.6

In modern Japan, river embankments have been constructed to mitigate flood risk; consequently, the current threshold for flooding is daily precipitation exceeding 100 mm [46]. However, because daily precipitation exceeding 100 mm is extremely rare, we used the frequency of days with precipitation exceeding 50 mm as a proxy for flood occurrence.

Our analysis indicates that the modern probability of fire remains highest between March and May. Similarly, the probability of daily precipitation exceeding 50 mm is highest between July and September. Despite the differences in data collection methods and geographic coverage, as well as the qualitative descriptions of each fire and flood record, the seasonal probabilities of fire and flood occurrence during the early 19th century appear remarkably consistent with present patterns.

4.3. Limitations and Uncertainty

Retrieving disaster records from old diaries and documents written by a single author involves the following uncertainties: (1) omission bias: disasters that actually occurred may not have been recorded; (2) document loss: old diaries and documents that recorded disasters may have been destroyed by war or subsequent natural disasters; (3) geospatial ambiguity: descriptions of locations may be unclear, or historical place names may have become defunct; (4) regional recording bias: authors prioritize different events based on local geography and culture; and (5) short-term missing records: the authors’ illness or personal circumstances may have caused them to miss recording some events.

To reduce the first and second uncertainties, researchers should retrieve as many old diaries and documents as possible from the same period. In addition, the use of probability mass functions, as demonstrated in this study, can help estimate occurrence patterns despite potential gaps. In addressing the third uncertainty, it is important to carefully cross-reference associated literature, but also allow contextual interpretation to some extent. Regarding the fourth uncertainty, it should be taken into consideration that historical records of flooding are often more prevalent in specific rural areas where flooding has occurred often since ancient days (e.g., river basins of the Kiso, Nagara, and Ibi Rivers) [8,10]. This fact indicates the importance of selecting target disasters that, as a consequence of the specific climate and topography, are likely to occur in the study area. Finally, regarding the fifth uncertainty, utilizing multiple historical sources is useful, although it introduces additional challenges.

5. Conclusions

We reconstructed the seasonal probability of fire and flood events in early-19th-century Kanazawa by utilizing disaster records extracted from the Kakuson Diaries. Our results demonstrate that fires were more likely to occur between March and June, driven by seasonal low humidity and strong winds, while floods were more likely to occur in June, July, and September. These historical patterns closely mirror current seasonal probabilities for fires and daily precipitation exceeding 50 mm, suggesting a long-term consistency in the region’s climate-driven disaster cycles.

While individual old diaries and documents provide high-resolution “historical dark data”, they are subject to uncertainties caused by omitted entries, the loss of archival material, ambiguous descriptions, unique local recording biases, and short-term missing records. To mitigate these limitations, future research should integrate records from multiple historical sources and focus on retrieving records for disaster types that are most probable given the local climate and topography. Combining such qualitative historical sources with the analysis of the seasonal probability of each disaster occurrence is a vital approach for understanding long-term environmental shifts and improving historical disaster reconstructions.